Abstract: Horizontal trend lasts for a short time, and the uncertainty of its direction changes is huge, so it becomes hard to forecast the direction of horizontal condition’s trend in Stock Market trend prediction. Based on the energy calculation of Horizontal window, BP neural network algorithm（WE-BPNN） is presented for predicting Horizontal window trend. Firstly, the division standard for short-term trend is given , on the basis of which this paper comes up with definitions of horizontal window. Then, by calculating the energy of K-line combination and moving average combination, two types of energy are merged into window energy. At last, leading the window energy into the direction of BP neural network to predict window trend. Because of hysteresis of energy’s influences on the trend, there is a case that energy accumulated while trend not changes, it will affect the accuracy of the trend prediction. Thus, basing on WE-BPNN neural network algorithm energy regulator is led into  BP neural network algorithm（EF-BPNN）, weights of window energy are dynamically adjusted for the trend prediction. On the Shanghai Stock’s data, the experimental results show that EF-BPNN algorithm has better performance.

Key words: energy window, K-line features, BP neural network, horizontal trend

摘要： 水平趋势持续时间短，方向变化的不确定性大，水平状态下趋势预测成为股市趋势预测的难点。基于水平窗口的能量计算，提出一种水平窗口趋势预测的BP神经网络算法（WE-BPNN）。算法首先给出短线趋势划分标准，在此基础上引入水平窗口定义；然后，通过对K线组合能量和均线组合能量进行量化计算，融合这两种能量得到窗口能量；最后，将窗口能量引入到BP神经网络预测窗口方向。由于能量对于趋势的作用具有滞后性，存在能量蓄而不发的情况，会影响到趋势判断的准确性，因而在WE-BPNN算法的基础上给出引入能量调节因子的BP神经网络（EF-BPNN）算法，动态调整窗口能量因子对于趋势预测的影响权重。在上证数据上的实验结果表明，EF-BPNN算法具有更好的性能。

关键词: 能量窗口, K线特征, BP神经网络, 水平趋势